Evaluation of Measures to Assess a Bank’s Credit Loss Experience

1. Evaluation of Measures to Assess a Bank’s Credit Loss Experience Slides prepared byKurt HessUniversity of Waikato Management School, Department of FinanceHamilton, New Zealand

2. Motivation Literature review Credit loss data Australasia Evaluation of CLE measures Potential proxies Correlations, lead/lag characteristics Reference levels Conclusions Topics Kurt Hess, WMS kurthess@waikato.ac.nz

3. Motivation • Stability and integrity of banking systems are of utmost importance to national economies • Credit losses, or more generally, asset quality problems have repeatedly been identified as the ultimate trigger of bank failures [e.g. in Graham & Horner (1988), Caprio & Klingebiel (1996)] • Entities in charge of prudential supervision and system stability thus need to understand drivers of credit losses in banking system Kurt Hess, WMS kurthess@waikato.ac.nz

4. Motivation • Very topical research area in the context of New Basel II Capital Accord • Basel II will allow use of proprietary models to determine required capital but these models & parameters require validation by supervisors • Need to understand potential procyclical effects which could endanger system stability Kurt Hess, WMS kurthess@waikato.ac.nz

5. Credit Risk & Basel II Basel on the Rhine RiverRetrieved from http://www.basel.ch 21 September 2004 Kurt Hess, WMS kurthess@waikato.ac.nz

6. Motivation • Methodological aspects particularly with regard to obtaining good data for this research have received scant attention • This presentation highlights some of the issues that were encountered when capturing a comprehensive credit loss history for Australian and NZ Banks (1980 – 2005) Kurt Hess, WMS kurthess@waikato.ac.nz

7. Motivation Methodological issues relate to . . . • Heterogeneity of reporting • Developed reporting typology to extract data along equivalent informational content • Choice of suitable proxies to measure credit loss experience (CLE) • Present results of an investigation on the properties of such CLE proxies Topic of Today’s Presentation Kurt Hess, WMS kurthess@waikato.ac.nz

8. Motivation Methodological issues relate to . . . (2) • Choice of appropriate explanatory variables • Explored characteristics / availability of data in Australasia & predictions by earlier research • Choice of suitable estimation models for highly unbalanced panel data Kurt Hess, WMS kurthess@waikato.ac.nz

9. Literature review Two main streams of research that analyse drivers of banks’ credit losses (or more specifically loan losses): • Literature with regulatory focus looks at macro & micro factors • Literature looks discretionary nature of loan loss provisions and behavioural factors which affect them Kurt Hess, WMS kurthess@waikato.ac.nz

10. Literature review Literature which explores macro and micro (bank specific) determinants of loan losses • Examples macro factors: • GDP growth, unemployment rate • indebtedness of households and firms • asset prices (real estate, share markets) Kurt Hess, WMS kurthess@waikato.ac.nz

11. Literature review • Examples of micro (bank specific) factors: • exposure to certain lending, collateral • portfolio diversification • (past) credit growth • net interest margins • efficiency Kurt Hess, WMS kurthess@waikato.ac.nz

12. Literature review • Behavioural hypotheses in the literature on the discretionary nature of loan loss provisions • Income smoothing: Greenawalt & Sinkey (1988) • Capital management: Moyer, 1990 • Signalling: Akerlof, 1970, Spence, 1973 • Taxation Management Kurt Hess, WMS kurthess@waikato.ac.nz

13. Literature review • Bank data in this literature typically sourced from third parties • Literature using commercial data providers:Cavallo & Majnoni (2001), Bikker & Metzemakers (2003) • Literature (partially) based on confidential data reported to regulators:Arpa et al. (2001), Keeton (1999), Quagliarello (2004) Kurt Hess, WMS kurthess@waikato.ac.nz

14. Literature review • Research based on original published financial accounts are rare, possibly due to very large effort to collect data Examples are • Pain (2003): 7 UK commercial banks & 4 mortgage banks 1978-2000 • Kearns (2004): 14 Irish banks, mostly early 1990s to 2003 • Salas & Saurina (2002): Spain Kurt Hess, WMS kurthess@waikato.ac.nz

15. Credit Loss Data Australasia • The database includes extensive financial and in particular credit loss data for • 23 Australian + 10 New Zealand banks • Time period from 1980 to 2005 • Approximately raw 55 data elements per institution, of which 12 specifically related to the credit loss experience (CLE) of the bank Kurt Hess, WMS kurthess@waikato.ac.nz

16. Credit Loss Data Australasia Sample selection criteria • Registered banks • Must have substantial retail and/or rural banking business • Exclude pure wholesale and/or merchant banking institutions Kurt Hess, WMS kurthess@waikato.ac.nz

17. Credit Loss Data Australasia Banks in sample AUSTRALIA: Adelaide Bank, Advance Bank, ANZ, Bendigo Bank, Bank of Melbourne, Bank West, Bank of Queensland, Commercial Banking Company of Sydney, Challenge Bank, Colonial State Bank, Commercial Bank of Australia, Commonwealth Bank, Elders Rural Bank, NAB, Primary Industry Bank of Australia, State Bank of NSW, State Bank of SA, State Bank of VIC, St. George Bank, Suncorp-Metway, Tasmania Bank, Trust Bank Tasmania, Westpac NEW ZEALAND: ANZ National Bank, ASB, BNZ, Countrywide Bank, NBNZ, Rural Bank, Trust Bank NZ, TSB Bank, United Bank, Westpac (NZ) Kurt Hess, WMS kurthess@waikato.ac.nz

18. Credit Loss Data Australasia The 1980 to 2005 data cover one ‘credit cycle’Australia Net write-offs as % of loans major banks Kurt Hess, WMS kurthess@waikato.ac.nz

19. Credit Loss Data Australasia The 1980 to 2005 data cover one ‘credit cycle’New Zealand Total stock of provision in banking system Kurt Hess, WMS kurthess@waikato.ac.nz

20. Credit Loss Experience of Australasian Banks Evaluation of CLE Measures

21. Principal Model CLEit credit loss experience for bank i in period t xit observations of the potential explanatory variables β(L) vector of polynomial in the lag operator associated with these explanatory variables uit random error term with distribution N(0,),  is variance-covariance matrix of it error terms q maximum lag of the dynamic component of the model Kurt Hess, WMS kurthess@waikato.ac.nz

22. Measuring CLE • Many proxies for a bank’s credit loss experience (CLE) are possible • Level of bad debt provisions, impaired assets, past due assets • Impaired asset expense (=provisions charge to P&L) • Write-offs (either gross or net of recoveries) • Components of above proxies, e.g. general or specific component of provisions (stock or expense) Kurt Hess, WMS kurthess@waikato.ac.nz

23. Measuring CLE Histogram of selected CLE proxies Extreme loss events of particular concern for credit risk management Median Pooled observations of Australian and NZ Banks 1980 - 2005 Kurt Hess, WMS kurthess@waikato.ac.nz

24. Measuring CLE Contemporaneous correlations between selected CLE proxies IAE_LN Imp. asset exp as % of loans IAE_NI Impaired asset expense as % net interest income IAE_GI Impaired asset expense as % gross interest income NW_LN Net debt write-offs as % of loans GW_LN Gross debt write-offs as % of loans RC_LN Recoveries as % of loans PRV_LN Provisions total as % of loans GE_LN General provisions total as % of loans SP_LN Specific provisions total as % of loans IA_A Impaired assets as % total assets PD_A Past due loans as % total assets GEE_LN Genl. provision expense as % of loans SPE_LN Spec. provision expense as % of loans Kurt Hess, WMS kurthess@waikato.ac.nz

25. Measuring CLE Lead / lagged correlations between selected CLE proxies Where the lead/lag correlation exceeds the corresponding contemporaneous value, one can say that the CLE proxy in the left column leads the proxy in the top row. Kurt Hess, WMS kurthess@waikato.ac.nz

26. Measuring CLELead-lag characteristic rooted in life cycle of bad debt provisioning Kurt Hess, WMS kurthess@waikato.ac.nz

27. Measuring CLE Modelling lag characteristic of write-offs: net write-off as a linear function of previous year impaired asset expense NW_LNit: Net debt write-offs as % of average loans of bank i in year tIAE_LNit: Impaired asset expense as % of average loans of bank i in year t Kurt Hess, WMS kurthess@waikato.ac.nz

28. Measuring CLE Modelling lag characteristic of write-offs: results Kurt Hess, WMS kurthess@waikato.ac.nz

29. Measuring CLE Modelling lag characteristic of write-offs: Interpretation of results • On average 1 Dollar in provisions expense is written down as follows: • Subsequent year 25 cts. • Year 2 30 cts. • Year 3 6 cts. • Year 4 14 cts. • This means only 75% of a year’s impaired asset expense is truly written off in the subsequent four years • Similar write-down patterns were found by Pain (2003) for UK major banks Kurt Hess, WMS kurthess@waikato.ac.nz

30. Measuring CLE Modelling lag characteristic of recoveries: Similar as previous results for write-offs • In theory, write-offs should mean losses with high degree of certainty • In practice, banks appear to interpret this differently • Across the sample cumulative bad debt recoveries as % of cumulative write-offs are 13.9% • These values vary significantly among banks(see following chart) Kurt Hess, WMS kurthess@waikato.ac.nz

31. Measuring CLE Cumulative debt recoveries as % of write-offs Kurt Hess, WMS kurthess@waikato.ac.nz

32. Measuring CLE: Reference levels • Reference levels (ratio denominator) to measure CLE • Literature typically uses levels of assets or loans (average of beginning and ending balance) • Can also consider income items like gross interest income, net interest income, total operating income Kurt Hess, WMS kurthess@waikato.ac.nz

33. Measuring CLE: Reference levels • It is found that balance sheet items have more desirable properties as reference levels • Main reasons are their magnitude & stability so CLE in numerator becomes major driver in derived ratio. Kurt Hess, WMS kurthess@waikato.ac.nz

34. Measuring CLE: Reference levels Balance sheet item growth (blue) is less volatile than changes in income items (red) Kurt Hess, WMS kurthess@waikato.ac.nz

35. Conclusions • Correlations between commonly used proxies rather weak • Only 75% of provision expense “turns into” write-offs • Write-offs do not seem definite as some banks subsequently recover up to a quarter of them Kurt Hess, WMS kurthess@waikato.ac.nz

36. Conclusions (2) • Choice of CLE proxy: None seems 100% ideal but ratios based on • Impaired asset expense (provision expense) still most preferable with best availability • Write-offs, while more certain, are too much delayed • Use assets or loans as a reference level Kurt Hess, WMS kurthess@waikato.ac.nz

37. Conclusions (3) • In summary, methodological issues related to modelling credit loss experience (CLE) may not be underestimated • Very important to select good proxies in calibrating risk models in the context of Basel II implementation Kurt Hess, WMS kurthess@waikato.ac.nz

38. Credit Loss Experience of Australasian Banks Back-up Slides

39. Basel II Pillars • Pillar 1: • Minimum capital requirements • Pillar 2: • A supervisory review process • Pillar 3: • Market discipline (risk disclosure) Kurt Hess, WMS kurthess@waikato.ac.nz

40. Basel II Pillars Pages in New Basel Capital Accord (issued June 2004) Kurt Hess, WMS kurthess@waikato.ac.nz

41. Pro Memoria: Calculation Capital Requirements under Basel II Unchanged Total Capital Credit Risk + Market Risk + Operational Risk  8% (Could be set higher under pillar 2) Significantly Refined Relatively Unchanged New Source: slide inspired by PWC presentation slide retrieved 27/7/2005 from http://asp.amcham.org.sg/downloads/Basel%20II%20Update%20-%20ACC.ppt , Kurt Hess, WMS kurthess@waikato.ac.nz

42. Basel II – IRB Approach Two approaches developed for calculating capital minimums for credit risk: • Standardized Approach (essentially a slightly modified version of the current Accord) • Internal Ratings-Based Approach (IRB) • foundation IRB - supervisors provide some inputs • advanced IRB (A-IRB) - institution provides inputs Kurt Hess, WMS kurthess@waikato.ac.nz

43. Basel II – IRB Approach • Internal Ratings-Based Approach (IRB) • Under both the foundation and advanced IRB banks are required to provide estimates for probability of default (PD) • It is commonly known that macro factor are the main determinants of PD Kurt Hess, WMS kurthess@waikato.ac.nz

44. Primer Loan Loss Accounting Kurt Hess, WMS kurthess@waikato.ac.nz

45. Primer Loan Loss Accounting Kurt Hess, WMS kurthess@waikato.ac.nz

46. Credit Losses and GDP Growth (New Zealand Banks) Provisioning/write-off behaviour correlated to macro factors Note: chart for NZ Bank sub-sample only Kurt Hess, WMS kurthess@waikato.ac.nz

47. Determinants of Credit Losses Macro Factors (1) Kurt Hess, WMS kurthess@waikato.ac.nz

48. Determinants of Credit Losses Macro Factors (2) Kurt Hess, WMS kurthess@waikato.ac.nz

49. Determinants of Credit Losses Bank Specific Factors (1) Kurt Hess, WMS kurthess@waikato.ac.nz

50. Determinants of Credit Losses Bank Specific Factors (2) Kurt Hess, WMS kurthess@waikato.ac.nz